Even as the Cancer Genome Atlas consortium continues to generate new data on breast cancer samples — including whole-genome and transcriptome sequences — some members of the team are finding ways to start applying the insights gained through exome sequencing and other analyses in clinical trial design and interpretation.
Earlier this week, TCGA researchers reporting in Nature outlined findings from exome sequence, microRNA sequence, gene expression, copy number, DNA methylation, and proteomic experiments done on as many as 800 matched tumor-normal samples from women with breast cancer.
Patterns in 466 cases — those profiled using everything but the protein assessment method — confirmed four breast cancer subtypes that were broadly consistent with gene expression-based classifications used previously. At the same time, though, the results provided greater resolution on the range of genetic glitches that occur within each subtype — details that are helping to better define and inspire treatment strategies for each group.
"I was very struck by the high correlation between mutations of specific genes only occurring in some specific subtypes," University of North Carolina, Chapel Hill, and Lineberger Comprehensive Cancer Center researcher Charles Perou, the study's corresponding author, told Clinical Sequencing News.
For example, he explained, researchers saw some recurrent mutations that appear to be specific to estrogen receptor-positive, luminal A tumors, which represent the most commonly diagnosed breast cancer subtype in the US. They also found marked similarities between difficult-to-treat breast cancers in the basal-like subtype and a type of tumor found in the ovary.
TCGA members are continuing to assess some of the tumor-normal sets included in the current study, Perou noted. Their goal is to analyze all 825 of those cases — and a few hundred more — using all six methods described in Nature this week.
They have also done whole-genome sequencing on around 50 of the tumors and RNA-sequencing on roughly 800 breast cancers. Studies describing those datasets have not yet been published but are anticipated in the not-too-distant future.
In the meantime, information gleaned from existing breast cancer data is being carefully mulled over by those interested in figuring out how to better define breast cancers at the time of diagnosis so that each patient receives the optimal treatment.
"From this study, there will be a wealth of new possible biomarkers that need to be evaluated in clinical trials," Perou said. "And many of the existing signatures that were tested were also further validated and will be tested in future clinical trials."
"The beauty now is that we have the framework and we know most of the players," he added. "We just need to take these into the clinical trial setting to see which ones are the most valuable predictors of drug responsiveness."
Though TCGA as a group is not directly involved in clinical trials of breast cancer, some members of the consortium belong to clinical trial groups that can apply the data both in new clinical trial design and for interpreting the results from past trials.
For instance, Perou is part of a clinical trial group called the Alliance that plans to retrospectively look at how hypotheses generated from the current TCGA breast cancer study correspond to past clinical trial outcomes — and to the genetic profiles of tumors collected for those trials.
Researchers are also mining the data to look for frequently altered genes and pathways within breast cancer subtypes that are already targeted in other cancers or which might inspire the adoption of altogether new treatments.
"This is a sort of roadmap for developmental therapeutics — drugs in development that aren't approved yet," co-lead author Matthew Ellis, head of medical oncology at the Washington University School of Medicine and director of that center's breast cancer program, told CSN.
For his part, Ellis is keen to see a move from retrospective analyses to genomics-forward medicine, in which "you have the genome in your hand and then it raises a therapeutic hypothesis which you address in that patient."
On that front, Ellis noted that some TCGA members have been working together to look at ways of potentially shifting cancer-related clinical trials organized through the National Cancer Institute's Clinical Trials Cooperative Group Program so that these trials incorporate sequencing and other 'omic approaches prospectively rather than retrospectively.
Because TCGA data is open to other researchers, those involved in the effort are optimistic that the breast cancer genetic profiles they are generating will inspire others to come up with new ways to tackle breast cancers as well.
The dataset generated for the current study includes exome sequences for 510 tumors from 507 women, along with matched normal exome sequences, plus miRNA sequence data for almost 700 cases. Between around 350 and 800 of the 825 tumor-normal pairs available were also assessed by protein profiling and/or array-based genotyping, expression, or copy number analyses.
When the researchers considered 466 tumors assessed by exome sequencing, miRNA sequencing, gene expression, copy number, and genotyping methods, they saw clusters coinciding with four main breast cancer subtypes: HER2-enriched, estrogen receptor-positive luminal A, estrogen receptor-positive luminal B, and basal-like.
The genetic and genomic patterns within the latter subtype, which includes "triple-negative" tumors lacking clear HER2, estrogen receptor, or progesterone receptor treatment targets, was especially intriguing, the team reported.
Rather than sharing features with the other three breast cancer subtypes, the basal-like tumors had mutation, gene expression, and copy number patterns that were far more similar to those described for serous ovarian tumors.
It had been known for some time that basal-like tumors had very different gene expression profiles than other breast cancers and shared at least two risk genes with ovarian cancers: BRCA1 and BRCA2. But the new study highlights the similarities between the basal-like tumors and their ovarian counterparts at other levels as well, suggesting that it might be possible to treat the two diseases using similar strategies.
Ellis explained that at the moment, both basal-like breast cancers and ovarian cancers are treated using chemotherapy. In the basal-like breast cancer case, these treatments are generally anthracycline-based. In ovarian cancer, on the other hand, clinicians have moved away from these anthracycline treatments in favor of platinum/taxol-based therapies with fewer side effects.
"Even if we did these randomized trials that we're beginning to talk about of an ovarian cancer regimen versus a breast cancer regimen in basal-like breast cancer," Ellis said, "and even if they were equivalent, the ovarian cancer regimen would win because it's safer."
The integrated genomic data for breast cancer also hints that there may be situations in which targeted treatments designed for other kinds of cancer could be effective for breast cancer.
For instance, the PIK3CA gene was one of just three genes mutated at more than 10 percent frequency across the breast cancer subtypes, hinting that it might be beneficial to treat some breast cancers with PI3-kinase inhibitors being developed in other cancers.
But it won't be until the appropriate trials are done that researchers will have a better idea of whether such inhibitors work against breast cancer or whether there are distinct PI3-kinase mutations in certain breast cancer subtypes that respond better to such treatment.
"In breast cancer, the only targeted therapies that are approved are estrogen receptor targeting drugs or endocrine therapies and HER2-directed drugs," Ellis said.
"We can see evidence of mutations that might sensitize to drugs that are used in other cancer types," he added. "The problem there is, of course, we still need to do the clinical trials to establish efficacy because it's not necessarily true that a mutation in, let's say, melanoma, would be equally druggable in breast cancer."
Moreover, in subtypes such as estrogen receptor-positive luminal A breast cancer, it may be possible to do more precise targeting since the new analysis indicates that mutations in PIK3CA and a few other genes tend to turn up at very specific sites in luminal A tumors.
"Luminal A [tumors] tended to have the fewest number of mutations but the highest number of recurrently mutated genes," Perou noted, "somewhat suggesting that when they have a mutation, it's really important."
In basal-like breast cancer, on the other hand, PI3-kinase activation seems to stem from the loss of players that would normally act as negative regulators of the pathway. That produces more broad-scale activation of the pathway, Ellis explained, suggesting that some tumors in that subtype might respond to more widely acting and potentially more toxic inhibitors of the pathway.
As additional layers of information are folded into TCGA's analyses, researchers expect to learn even more about the range of features found in each subtype and ways in which they impact treatment success.
For the HER2-enriched breast cancers, for instance, adding in protein profiling data has already helped to define subsets of clinical HER2-enriched cases that have especially high levels of HER2, phosphorylated HER2, and show other signs of HER2/EGFR pathway activation.
"We can see that there are at least two kinds [of HER2-enriched breast cancers]," Perou said. "Even within those called [HER2] 'positive,' we can see that there's a range of expression from very high to medium-high. And that includes a relative difference in phosphorylation status."
Again, more studies and clinical trial are needed to see if the tumors that have protein profiles that are indicative of more pronounced HER2/EGFR pathway activation also respond better to HER2-targeted treatments such as trastuzumab, marketed as Herceptin by Roche/Genentech, or the small molecule inhibitor lapatinib, GlaxoSmithKline's Tyerb.
"We need to test this to see if it actually does make a response difference," Perou said. "We don't know that today, but it's certainly a hypothesis and it's certainly a testable hypothesis."